11.07.2015 Views

Fire Detection Algorithms Using Multimodal ... - Bilkent University

Fire Detection Algorithms Using Multimodal ... - Bilkent University

Fire Detection Algorithms Using Multimodal ... - Bilkent University

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

CHAPTER 6. WILDFIRE DETECTION 100Universal Predictor(x,n)for i = 1 to M dow i (0) = 1 , InitializationMend forŷ u (x, n) = ∑ i v i(n)D i (x, n)if ŷ u (x, n) ≥ 0 thenreturn 1elsereturn -1end iffor i = 1 to M dov i (n + 1) =exp(− 1 2c l(y(x,n),D i(x,n)))∑j exp(− 1 2c l(y(x,n),D j(x,n)))l(y(x, n), D i (x, n)) = [y(x, n) − D i (x, n)] 2end forFigure 6.6: The pseudo-code for the universal predictoralgorithms are linearly combined similar to Eq. 6.9 as follows:ŷ u (x, n) = ∑ iv i (n)D i (x, n) (6.37)where the weights, v i (n), are updated according to the ULP algorithm, whichassumes that the data (or confidence values D i (x, n), in our case) is governed bysome unknown probabilistic model P [74]. The objective of a universal predictor isto minimize the expected cumulative loss. An explicit description of the weights,v i (n), of the ULP algorithm is given as follows:v i (n + 1) =exp(− 1 2c l(y(x, n), D i(x, n)))∑j exp(− 1 2c l(y(x, n), D j(x, n)))(6.38)where c is a normalization constant and the loss function for the i-th decisionfunction is:l(y(x, n), D i (x, n)) = [y(x, n) − D i (x, n)] 2 (6.39)The constant c is taken as 4 as indicated in [74]. The universal predictor basedalgorithm is summarized in Fig. 6.6.We also implemented the Weighted Majority Algorithm (WMA) as the decisionfusion step [47]. The WMA is summarized in Fig. 6.7 [58]. In WMA,as opposed to our method, individual decision values from sub-algorithms are

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!